Skip to main content
Version: 2026.3

Embeddings (Experimental)

caution

This is an experimental feature. Configuration, data format, index layout, and APIs may still change in breaking ways between releases.

Embeddings turn the content of your elements into vectors and store them in the Generic Data Index. Searches then match by meaning instead of by words, and near-identical elements can be found without an exact match. A data object is embedded from its composed text, an image asset from a thumbnail and an asset text or document from its extracted text.

Configure everything in Pimcore Studio under System > Embedding Configuration.

Embedding Configuration in Pimcore StudioEmbedding Configuration in Pimcore StudioEmbedding Configuration in Pimcore Studio

The feature is off by default and calls an external inference service you provide — the bundle ships no model. See Getting Started to turn it on.

Features​

  • Semantic search: queries match by meaning instead of exact words ("British roadster" finds an MG even if no field contains "roadster"), including multilingual queries. See Semantic Search.
  • Image search: text queries against image assets ("yellow van"), no tagging required.
  • Duplicate detection: exact duplicates via a content hash, near-duplicates via vector and MinHash similarity, for objects and images. See Duplicate Detection.
  • Vectors are reused, not recomputed. Each stored vector carries an identity key derived from the model and the exact input, so re-indexing unchanged content costs no inference calls, see Caching.
  • Generation never blocks a save. It runs through Symfony Messenger. If the inference service is unavailable, the element is indexed without vectors, marked for backfill, and repaired later.
  • Compare models on your own data with the Evaluation Tool before you commit to one.

Pages in this chapter​

  • Getting Started: from a freshly installed bundle to working semantic search.
  • Configuration: every setting in the Studio editor, and the deploy-time YAML that has no editor.
  • Caching: how vector reuse works, when vectors are regenerated, and how to bypass it.
  • Inference Service: hosting options, the HTTP contract, and the serving invariants.
  • Defaults and Benchmarks: which model, text format and compression the bundle defaults to, and the measurements behind each choice.
  • Evaluation Tool: benchmark search and duplicate-detection quality against curated label files.
  • Customization: write your own provider or register your own usage.
  • Progress and Cancellation: follow a re-embed run and cancel it.

The Embedding Configuration screen is admin-only; it has no dedicated permission, see User Permissions.